search

dongwonlee-lab / tsim

0 stars 0 forks source link

number of associated SNPs in tsim Cohorts GWAS than meta-analysis of GWASs #1

Open Manonanthini opened 3 months ago

Manonanthini commented 3 months ago

Hi, The two-stage imputation strategy is great. Thank you for the method! I have tried tsim in two differently genotyped independent Cohorts after the first imputation using the same reference panel and GWAS after merging them using hq SNPs followed by the second stage imputation using the Topmed r3 reference panel. I also performed GWAS on them independently and combined them using meta-analysis. I found less number of SNPs in the tsim merged GWAS compared to the meta-analysis of GWAS results of those cohorts. I found almost twice the number of SNPs in meta-analysis than merged GWAS. What should be the reason for that? How can I rectify it?

Regards Mano

Dongwon-Lee commented 3 months ago

Hi Mano, thank you for your interest in our TSIM method! For the r3 TOPMED reference panel, you will need to use a lower r^2 cutoff for hqSNP to have sufficient coverage. As noted in the README, we have found that 0.98 should be used instead of 0.99. In fact, you can still get a comparable accuracy with an r^2 cutoff as low as 0.97 based on our new calibration analysis. We are in the process of revising the manuscript to reflect this. Stay tuned!

Dongwon

Manonanthini commented 3 months ago

Hi, Thanks for the reply. I need a clarification regarding ER2 value. Do we have to keep the threshold as 0.9 or can it be reduced further down? If so, what should be the minimum value to consider for ER2 ? Also for hq SNPs, is ER2 >0.9 & R2>0.97 or ER2 >0.9| R2>0.97 is used in the code? Regards Mano

On Thu, 13 Jun, 2024, 22:24 Dongwon Lee, @.***> wrote:

Hi Mano, thank you for your interest in our TSIM method! For the r3 TOPMED reference panel, you will need to use a lower r^2 cutoff for hqSNP to have sufficient coverage. As noted in the README, we have found that 0.98 should be used instead of 0.99. In fact, you can still get a comparable accuracy with an r^2 cutoff as low as 0.97 based on our new calibration analysis. We are in the process of revising the manuscript to reflect this. Stay tuned!

Dongwon

— Reply to this email directly, view it on GitHub https://github.com/dongwonlee-lab/tsim/issues/1#issuecomment-2166700390, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACPEQJEO6GNSTI6LE5QXPYDZHH5WLAVCNFSM6AAAAABJJCQDY2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCNRWG4YDAMZZGA . You are receiving this because you are subscribed to this thread.Message ID: @.***>

Dongwon-Lee commented 3 months ago

Currently we use "AND". Since ER2 is calculated based on the directly genotyped data, it is not affected by the changes in the reference panel. Therefore, the ER2 threshold should remain the same.

Manonanthini commented 3 months ago

Hi,Thank you for the clarification. As the ER2 threshold 0.9 is very high and you have used 'AND' with R2>0.97, a lot of directly genotyped SNPs will be deleted from the dataset if I understand correctly. Shall i use 0.7 or 0.8 as threshold for ER2 to include more directly genotyped SNPs into the dataset?

Regards Manonanthini

On Fri, 14 Jun, 2024, 17:19 Dongwon Lee, @.***> wrote:

Currently we use "AND". Since ER2 is calculated based on the directly genotyped data, it is not affected by the changes in the reference panel. Therefore, the ER2 threshold should remain the same.

— Reply to this email directly, view it on GitHub https://github.com/dongwonlee-lab/tsim/issues/1#issuecomment-2168258193, or unsubscribe https://github.com/notifications/unsubscribe-auth/ACPEQJDOREQ6PAVDP52G42TZHMCYFAVCNFSM6AAAAABJJCQDY2VHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDCNRYGI2TQMJZGM . You are receiving this because you are subscribed to this thread.Message ID: @.***>